We performed an intercomparison of three eddy-resolving ocean reanalyses (FORA-WNP30, GLORYS12V1, and BRAN2020) for the western North Pacific over the period 1993--2016. While all three products exhibit high consistency in large-scale surface mean fields due to data assimilation, notable discrepancies are found in mesoscale variability and subsurface structures, reflecting differences in model configuration and assimilation schemes. FORA-WNP30, employing a four-dimensional variational (4D-VAR) scheme, exhibits high spatiotemporal variability in sea surface height (SSH). However, its sea surface salinity (SSS) variability is strongly dampened by a short restoring timescale. GLORYS12V1 retains intense coastal SSS variability due to the absence of strong surface restoring. However, it exhibits lower eddy kinetic energy levels compared to the others, likely due to dissipative viscosity parameterization. BRAN2020 shows systematic differences east of Japan, including reduced SSH variability and lower temporal correlations. These are attributed to its simplified bathymetry (capped at 5000 m), which eliminates topographic steering effects and alters eddy propagation speeds. Based on these findings, we will discuss how horizontal resolution is not the sole determinant of performance; bathymetry, subgrid-scale parameterizations, and boundary conditions play equally critical roles in reproducing regional ocean dynamics.